Re: [PATCH v0] fast-import: Add drop command

[Vitor Antunes <vitor.hda@gmail.com>]

On Sat, Sep 24, 2011 at 10:19 PM, Dmitry Ivankov <divanorama@gmail.com> wrote:
> On Sun, Sep 25, 2011 at 1:37 AM, Jonathan Nieder <jrnieder@gmail.com> wrote:
>> Thanks.  I must have missed the earlier discussion.  What are the
>> semantics of this command and its intended purpose?
> My guess is that if fast-import is used to manage a set of "remote"
> branches, it should be able to delete branches. Then, it should
> be allowed to do non-fastforward updates too (--force). Why can't
> it just ignore branches deletion (considering --force)?

I started by using --force, but I did not want to completely disable
these checks. The idea of the drop command is to add support to the
exceptions that require non-fastforward updates.

> Random thoughts:
> 1. once 'drop' is executed, fast-import can't tell if the branch was
> actually deleted. And moreover any attempt to read this branch
> head becomes illegal (either it's missing in .git or fast-import is
> instructed to use a dropped branch).
> 2. 'reset' command is a bit like proposed 'drop' but it never deletes
> a branch ref. Consider following imports:
> 1) import branch topic
> 2) reset topic
> 3) import branch topic2 starting at topic (incorrect import)
> If 1-3) is done in one fast-import process, the error is reported.
> If 3) is done separately, it succeeds but the result is strange:
> topic2 isn't started from scratch but from old "erased" topic.
> So, maybe, reset should be fixed to erase branches on --force.

I think you are not considering the possibility that checkpoints could
have been done along the way. I use them frequently to be able to
analyse branches with diff-tree. As soon as a checkpoint is done,
update-branches will issue an error (commit A is not part of branch A').

> One more scenario is:
> 1) import topic
> 2) reset topic
> 3) import topic
> If 1-3) go together - no error
> If 3) goes separate - no error, but non-fastforward update.
> Much more harmless, but still may look strange.

Not exactly true if there is a checkpoint done after step 1.

My scenario is:

1) import topic
2) checkpoint
3) diff-tree and processing
4) exit if processing returns ok
5) reset topic to another HEAD
6) goto 1)

-- 
Vitor Antunes